Dear Maintainers,
please note that
* nvidia-kernel-*: [i386] Drop support for building amd64 kernel modules.
from 340.76-5 also hits 340.76-5~bpo8+1 from jessie-backports, which
might be considered a regression. In my case, I was running 340.76-2
(manually "backported" to jessie from unstable, meaning I just built
it there) and was hit (dkms module build failing) when I tried to upgrade
that to 340.76-5. At the time I didn't identify the culprit was just this
change.
Is there any significant technical issue that made the change necessary?
I'm asking because I just added amd64 back to ARCH_DEPS_i386, rebuilt
340.76-5~bpo8+1 with that single change, and the resulting nvidia-kernel-dkms
package compiled successfully against my kernel. That allowed me to finally
upgrade to 340.76-5~bpo8+1 and in turn kernel 4.1.8, and the result is
running without issues. I even managed to build 340.93-2 the same way
(again just adding back ARCH_DEPS_i386 ?= amd64 in debian/rules) with the
additional complication of having to also backport glx-alternatives-0.6.0
(presumably a proper backport by a more knowledegable person than me would
work around that by packaging against the jessie 0.5.1 versions of these).
I'm running jessie i386 userspace on a amd64 kernel built and installable
on i386 (make-kpkg --arch=amd64 --cross-compile=-) and don't plan to
change that any time soon (the installation has to be able to boot and
run with a i386 kernel as well, so multiarch is no option, and as long
as there is no easy and proven safe way for sidegrading i386 to amd64, I
can't enable multiarch on production systems anyway). I was under the
impression that this (i386 userspace on amd64 kernel with no additional
multiarch whatsoever) was officially supported for jessie (given the
presence of an accordingly built linux-image-amd64 in the i386 part of
the repo). It would be nice to have support for this use case continued
without jumping through hoops. I don't even know how I would attempt
building e.g. nvidia-kernel-dkms:amd64 from the nvidia-graphics-drivers
source on my i386 userspace.
So please correct me if I'm wrong with any of my conclusions. I know
the bug was closed by the OP, so do what you want with this message.
I just wanted to provide an additional PoV on the problem, add another
voice telling you the feature you removed was still in use out there,
and would like you to consider a reopen on wishlist severity.
Thanks & Regards,
Andre.
--
Cool .signatures are so 90s...
-> Andre Beck +++ ABP-RIPE +++ IBH IT-Service GmbH, Dresden <-
